DeepSeek V3 API pricing
As of October 21, 2026, DeepSeek V3 costs $0.2574 per million input tokens and $1.0287 per million output tokens through OpenRouter.
DeepSeek V3 benchmarks and specs · Every DeepSeek model's price
- Input, per million tokens$0.2574
- Output, per million tokens$1.0287
- Cached input, per millionNo cache price listed
- Context window164K tokens
- Max output16K tokens
- Output speedNot measured
What a request costs
Three fixed token mixes at the list price, without cache or batch discounts. For a whole month, use the calculator.
| Workload | Tokens | One request | 1,000 requests | Fits the context window |
|---|---|---|---|---|
| Chat turn | 1K input + 500 output tokens | $0.0008 | $0.772 | Yes |
| Repository review | 50K input + 3K output tokens | $0.016 | $15.96 | Yes |
| Agent loop | 200K input + 10K output tokens | $0.062 | $61.77 | No, it would need splitting |
| Solved coding task | One SWE-bench Verified attempt, divided by the share that succeed | Not computable | Not computable |
Price history
We have recorded 3 price changes since we started reading the price daily on October 15, 2026.
- October 15, 2026 First recorded: $0.32 input, $0.89 output
- October 16, 2026 $0.89 to $1.0287 output, $0.32 to $0.2574 input
- October 17, 2026 $1.0287 to $0.89 output, $0.2574 to $0.32 input
- October 20, 2026 $0.89 to $1.0287 output, $0.32 to $0.2574 input
Cheaper models as good as DeepSeek V3
Models within two points of DeepSeek V3 or above it on the Vibe Coding Index, with a lower output price. Cheapest first.
- Qwen3.8 Flash$0.47 Coding 87.4 against 70.2 · 54% cheaper output
- Claude Haiku 5.5$0.50 Coding 75.9 against 70.2 · 51% cheaper output
- GPT-6 Luna$0.50 Coding 71.4 against 70.2 · 51% cheaper output
- GLM 5.3 Flash$0.50 Coding 71.0 against 70.2 · 51% cheaper output
- DeepSeek V4.1 Flash$0.525 Coding 77.2 against 70.2 · 49% cheaper output
DeepSeek V3 pricing questions
- How much does DeepSeek V3 cost per million tokens?
- $0.2574 per million input tokens and $1.0287 per million output tokens, as listed on OpenRouter on October 21, 2026.
- Does DeepSeek V3 support prompt caching?
- OpenRouter lists no cached input price for this model, so assume every input token is billed at the full rate.
- Is there a cheaper model as good as DeepSeek V3?
- Qwen3.8 Flash scores 87.4 on the Vibe Coding Index, within two points of DeepSeek V3 or above it, and costs $0.47 per million output tokens.
- Has the price of DeepSeek V3 changed?
- Yes, 3 times since October 15, 2026; the dates are listed on this page.